Code Katas and Evidence

Use the offline lab package and guided sequence. Its six learner functions replace open-ended coding prompts with executable contracts and an independent public checker.

Kata	Before coding	What the check catches
posterior	Expected count table	Reversed conditioning and zero-probability evidence
retry_summary	Terminal paths or indicators	Counting unattempted requests and final backoffs
defect_pmf	Uniform unordered samples	Replacement mistakes, impossible counts, empty sample
wilson_interval	Supplied formula and endpoint cases	Wald substitution, out-of-range endpoints, roundoff at zero
exact_coverage	Weighted sum over possible counts	Unweighted averaging of counts and hardcoded .95
standardized_mean	Target workload distribution	Different comparison weights and missing groups

Work in starter.py; run python check.py --module starter. Use the reference implementation after attempting a derivation, and record any consultation. The simulation harness is supplied so learners can focus on models rather than CSV plumbing.

For a debugging exercise, temporarily replace your retry expected-attempt count with the maximum allowed count. Predict the failing case, run the checker, then repair the change. Repeat with a binomial PMF substituted for the without-replacement experiment. Explain why a test of the mean alone would miss the latter error.

Completion includes the written assessment, reproducible reports for two seeds, and a changed transfer case. Code checks cannot establish that a production population is independent or that its workload matches the fixture.